Revert CoVe contexts to full payloads: preserve all vector store metadata

**FIX:**

Line 292: Use full payloads from final_results, not filtered contexts
- Was: contexts=contexts (only id, text, doc_title, url, retrieval_score)
- Now: cove_contexts = [payload for _, payload, _ in final_results]
- Preserves ALL vector store metadata (position, total_chunks, metadata dict, etc.)

**Why this matters:**
- Vector store payloads contain additional fields beyond our filtered contexts
- CoVe citation_injector and verifier may need these metadata fields
- Filtering to only 5 fields could break CoVe's functionality
- Original code was correct - passed full payloads

**Example lost fields:**
- position (chunk position in doc)
- total_chunks (total chunks in doc)
- metadata dict (custom fields like doc_type, etc.)
- Any reranker-added fields (local_rerank_score, etc.)

Reverts to original implementation that preserves all payload data.

Author: Floressek

src/ragx/api/routers/eval.py

@@ -288,11 +288,14 @@ async def pipeline_ablation(
     if request.cove_enabled:
         start_cove = time.time()
 
-        # Run CoVe verification (reuse contexts built earlier)
+        # Prepare contexts for CoVe (full payloads with all metadata)
+        cove_contexts = [payload for _, payload, _ in final_results]
+
+        # Run CoVe verification
         cove_result = cove_enhancer.verify(
             query=query,
             answer=answer,
-            contexts=contexts,  # Already built as List[Dict] from final_results
+            contexts=cove_contexts,
         )
 
         # Update answer if corrections were made